Simplify the Development of Efficient Software Applications with Parallel Computing Cluster

Parallel computing cluster has emerged as a cost effective means to tackle computationally intensive problems in finance and engineering. Clustertech Parallel Environment (CPE) is a software suite simplifying the distribution of C/C++ workloads on a cluster to improve computation speed significantly.

Simplifying Parallel Software Development in C/C++

A major hurdle associated with migrating from serial programming to parallel programming is that the latter requires sophisticated synchronization and data transportation among parallel processes. CPE provides a standard platform for data manipulation in parallel programs, hiding the details of parallel synchronization and communication operations. Developers can focus on the core functionalities specific to their application, shorten the overall development time.

Enhanced Cluster Resources Sharing and Utilization

CPE allows users to launch a parallel application securely and transparently from a workstation through Local Area Network or over the INTERNET using its proprietary ParaConnect technology. The tight integration of CPE with cluster resources management systems, such as PBS Professional and Windows HPC Server, allows cluster to be effectively shared among multiple users.

Automated Distribution of Batch Jobs

A common use case across many parallel software applications is distributing a batch of jobs into a computer cluster to speed-up the calculation. Using the Job Distribution (JD) library of CPE, the jobs can be arranged to execute on a compute cluster with minimal programming effort. The JD library comes with the important features of dynamic load balancing and fault tolerance, ensuring the robustness of the parallel programs.

Optimized Libraries for Monte Carlo and Finite Difference Methods

Monte Carlo (MC) and Finite Difference (FD) methods are the two most common numerical procedures employed in financial and engineering applications. Both methods are computational intensive and can be efficiently speed-up using computer cluster. CPE provides optimized domain specific libraries for MC and FD methods to greatly reduce the development time of parallel codes and/or facilitate parallelization of existing application employing these two numerical procedures.

"The Clustertech Parallel Environment (CPE) is a software platform which facilitates the development, deployment and execution of parallel applications, unifying and greatly simplifying the process. The CPE core facilitates the transfer of data between machines and can be used to simplify the development of parallel software."

The MC library facilitates the development and porting of MC simulations. Users need only provide the code which generates a single path in the MC simulation, and choose among several path generation, statistics consolidation and termination policies. The MC library supports antithetic variables, quasi Monte Carlo simulation, American options, saving of paths, etc. In most MC applications, linear scaling (close to N times speedup on N processors) can be achieved.

The FD library facilitates the development and porting of finite difference codes for solving initial value problems of linear homogeneous partial differential equations (PDEs). Users only need to specify the PDE terms, free from the hassles of manipulating matrix elements. The FD library supports Implicit, Explicit, Crank-Nicholson and 4 variants of Alternating Direction Implicit (ADI) schemes (Douglas, Craig-Sneyd, Modi_ed Craig-Sneyd, Hundsdorfer-Verwer). Standard boundary conditions (including Dirichlet, Neumann and second order derivative boundary conditions) and standard discretization schemes (including centered and upwind discretization) are also supported.